The Census Bureau just published a working paper on using "data from one survey (typically a smaller survey with a rich set of items) to train a machine learning model to predict an outcome of interest." They also provide an example modeling "a variable, whether or not a housing unit had an air conditioning unit from the American Housing Survey to the American Community Survey."
Details and data:
Cross-Survey Modeling: Fusing Data from Multiple Data Sources to Enhance Multi-Dimensional Measures
Accelerating Entropy Balancing Sample Weighting
| Census.gov |
remove preview |
|
| Accelerating Entropy Balancing Sample Weighting |
| New optimization routines described and implemented to quickly weight large samples using Entropy Balance calibration. |
| View this on Census.gov > |
|
|
------------------------------
Beth Jarosz
------------------------------